Control System Synthesis by Root

Locus Method
WALTER R. EVANS

the effect of introducing additional open-loop poles and zeros.

In Evans's own words, "[the] particular advantage [of the root
locus method] is in retaining at all times a clear picture of the
relationships between the overall network parameters and the
parametersof individualcircuit elements."
G(s) _
KN(s)
The root locus quickly became a favorite tool of control en(1)
- D(s) + KN(s) ,
gineers, whose intuition was enriched by its clear visualization
where K N(s) / D (s) is the open loop transfer function and K is of feedback system properties. It has had a lasting impact on
the gain. The closed-loop zeros are the zeros of N(s) and the engineering education. Every present-day undergraduate control textbook includes a detailed discussion of the root locus
poles are the roots of the equation
method. The root locus is still a very effective way to quickly
(2) assesshowthepropertiesof a systemchangewiththegain.Every
D(s) +KN(s) = 0
control systems software package offers root loci as a standard
In the late 1940s it was still a tedious task to compute roots of
tool. Today, we still plot root loci but the graphical construcpolynomials. As an extension of a technique he had developed
tion is rarely used any more, because using efficient software
earlier in [2], and using homotopyideas,Evanspresentedin this
packages one can easily compute the roots with much greater
1950paper a clevermethodto findthe locus of roots of (2) when
accuracy.
the gain varies. He observed that the roots of (2) approach the
Root locus ideas contributed to the development of more adroots of D(s) and N(s) as K ~ 0 or 00. For small gains the
vanced feedback system concepts, including multivariable root
closed-looppoles are close to the open-looppoles, and for large
locus generalizations and high gain propertiesdisplayedby root
values of the gain deg N of the closed-loop poles migrate to
locus asymptotes. The fundamental property that closed-loop
the zeros and the remaining deg D - deg N closed-loop poles
poles tend to system zeros as the gain becomes very high is the
go towards infinity in a symmetric star configuration. Evans
essence of the modem concept of "zero dynamics," not only
developedsimplemethodsto sketchthe root loci based on initial
in linear, but also in a large class of nonlinear systems [1]. This
directions, asymptotes, multiple roots, and behavior close to
propertypointsto the obstaclesto achieving highperformancein
multiple roots.
systemswith unstablezero dynamics(non-minimum phase sysThe presentation of the method in the paper focuses on the
tems). Feedback limitationswere analyzed using the root locus
graphical construction. The paper even describes how to make
asymptotes of optimal systems in [3]. A design which assigns
a simple "spirule" (Fig. 8 of the paper) to facilitate the conthe rates with which roots move along their loci was developed
struction. Many of the basic root loci rules as they are usually
in [4]. These and similar recent results show that the impact
elaboratelylisted and emphasized in modem textbooksmay be
of the root locus method was broader than just being a useful
found in this paper, but they are introduced informally, often
graphical tool.
throughan example. They includethe start and end points of the
loci, their asymptotes, portionson the real axis, and the presence
of break-away points.
REFERENCES
The paper demonstrates convincingly how root loci are used
to establish importantpropertiesof closed-loopsystemssuch as [1] C.I. BYRNES AND A. ISIDORI, "A frequency domain philosophy for for nonlinear systems," Proc. 23th IEEE Con! Decision and Control, pp. 1569their responseas determinedby the dominantclosed-looppoles.
1573, Las Vegas, NV, 1984.
It does not stop at this, but also illustrateshow root loci may be [2] W.R. EVANS, "Graphical analysis of control systems," Trans. Amer. Inst.
Electric. Engineers, 67:547-551,1948.
usedto designcorrectivenetworksby anticipating and analyzing

MUCH insight into the properties of a single-input singleoutput linear system can be obtained from a plot of the poles
and zeros of its transfer function. For a single-loop system, the
closed-looptransfer function is given by

107

[3] H. KWAKERNAAK AND R. SNAN, "The maximally achievable accuracy of

linearoptimal regulators andlinearoptimalfilters," IEEETrans. Aut.Contr.,
17:79-86,1972.

[4] A. SABERI ANDP. SANNUTI, "Time-scale structure assignment in linearmultivariable systems usinghigh-gain feedback," Internat. J. Contr., 49:21912213,1989.

H.K.

108

Control System Synthesis by Root

Locus Method
WALTER R. EVANS
MEMBER AlEE

Synopsis: The root locus method determines all of the roots of the differential
equation of a control system by a graphical
plot which readily permits synthesis for
desired transient response or frequency
response. The base points for this plot
on the complex plane are the zeros and poles
of the open loop transfer function. which are
readily available. The locus of roots is a
plot of the values of which make this
transfer function equal to -1 as loop gain is
increased from zero to infinity. The plot
can be established in approximate form by
inspection and the significant parts of the
locus calculated ac:cura.te1y and quickly by
use of a simple device. For multiple loop
systems. one solves the innermost loop
first, which then permits the next loop to be
solved by another root locus plot. The
resultant plot gives a complete picture of
the system, which is particularly valuable
for unusual systems or those which have
wide variatiODs in parameters.

HE root locus method is the result of

an effort to determine the roots of the
differential equation of a control system
by using the concepts now associated
with frequency response methods,' The
roots are desired, of course, because they
describe the natural response of the system. The simplifying feature of the
control system problem is that the open
loop transfer function is known as a
product of terms. Each term, such as
l/(l+Ts), can be easily treated in the
same manner as an admittance such as
1/ (R +jx). I t is treated as a vector in the
seose used by electrical engineers in
solving a-c circuits. The phase shift and
attenuation of a signal of the fonn ~It
being transmitted is represented by
l/(l+Ts) in which. in general is a complex number. The key idea in the root
locus method is that the values of
which make transfer function around the
loop equal to -1 are roots of the differ..
ential equation of the system.

The opening section in this paper,

Background Theory, outlines the over-all
pattern of analysis. The following section on Root Locus Plot points out the
great usefulness of knowing factors of the
open loop transfer function in finding the
roots.
The graphical nature of the method
requires that specific examples be
used to demonstrate the method itself
under the topics: Single Loop Example,
Multiple Loop System, and Corrective
Networks. The topic Correlation with
Other Methods suggests methods by
which experience in frequency methods
can be extended to this method. The
topic Other Applications includes the
classic problem of solving an nth degree
polynomial. Finally, the section on
Graphical Calculations describes the key
features of a plastic device called a
"Spirule", which permits calculations to
be made from direct measurementon the
plot.

Background Theory
The over-all pattern of analysis can be
outlined before explaining the technique
of sketching a root locus plot. Thus
consider the general single loop system
shown in Figure 1.
Note that each transfer function is of
the form KG(a) in which K is a static gain
constant and G(.) is a function of the
complex number. In general, G(a) has
both numerator and denQ!l1inator known
in factored fonn. The values of which
make the function zero or infinite can
therefore be seen by iDspectioD and are

called zerosand poles respectively. The

closed loop transfer functioD can be expresseddirectly from Figure 1 as givenin
equation 1

~(.) _

'f

KIGI'(.)

l+K".G,.(.)K~JJ(.}

(1)

The problem of finding the roots of the

differential equa.tion here appears in the
form 'of finding values of which make
the denominator zero. After these values
arc determined by the root locus method,
the denominator can be expressed in
factored form. The zeros of the function
So/Bf can be seen from equation 1 to be the
zeros of G_{,) and the poles of GJJ(- ) .
The function can now be expressed as
shown in equation 2

~(.) _Xca.,(1-a/9,1)(1-_/fJ.1)
If

(2)

(1-./'1)(1-./,.) ....

The constant K, and the exponent ~

depend upon the specific system but for
control systems 'Y is.often zero and K, is
often 1.
The full power of the Laplace Transform' or an equivalent method now can
be used, The transient response of 'the
output for a unit step input, for example,
is given by equation 3

1-,.

(Jo(t) lie 1- ~ A. tern

"_1

(3)

The amplitude A, is given by equation

4

Ai_[!.o'. (.)(l-_/ri) 1.r,

(4)

The closed loop frequency response, on

the other hand, eau be obtained by substituting _,.j", into equation 2. For-

tunately, the calculation in finding At or

9o/8,(jfA') involves the same problem of
multiplying vectors that arises in making
a root locus plot, and can be calculated
quickly from the resultant root locus
plot.
Paper 50-11. reco.lllmea4.d by tbe AIBB Feedback.
CODtt'ol Syateaas ColDDIittee aDd approved by tbe

AIBB Tedualcal Prosram Committee for praentadOD at tile AIBE WiDter aeaera! Keetillit New
York, N. Y. lal1uaryao-PebntarT8, 1950. MaaulUipt lablllltted Novber 15. 1948; made avail
able f9J' priDtiDI November' 12. 1949.
W ALTB. R.. VAllI is with North AmericaD Aviation, IDC.. Done,.. Calli.

The author wilhea to UJ)retj his appredatioa lor

the auiataaee ...... by . . fellow wwker., IC. ll.
Jacksoa aud R.. M. OJborD, ia the preparation of this
paper. la particuJar, Mr. Ostom coatrlbuted the
circuit aaalytil Rumple.

109
Reprinted with permission from Transactions of the American Institute of Electrical Engineers, Walter R. Evans, "Control
System Synthesis by Root Locus Method," Vol. 69, 1950, pp. 66-69.

'.

'-

J.

Figure 1 (left). General blacle dilgram

Figure 3 (right). Singl.loop root locus

If

Root Locus plot

The open loop transfer function is
typically of the form given in equation 5.
KIAG~(.)K,GB(.)
K(l+T~)
II:

s(1

(cral+"'It]

+T._) 1(_+",) t+~.2]

(5)

The parameters such as T1 are constant

for a given problem, whereas assumes
many values; therefore, it is convenient
to convert equation 5 to the form of

equation 6.
K_G,,(.)K~G~(.)

KeltT2 +_) TI[cral +wi 2]

- .(1/T1+.)T1 [(.+cr3+j~) (.+CTI-j~l]
(6)

The poles and zeros of the function are

plotted and a general value of is assumed as shown in Figure 2.
Note that polesare represented as dots,
and zeros as crosses. All of the complex
tenns involved in equation 6 are represented by vectors with heads at the general point _ and tails at the zeros or poles.
The angle of each vector is measured with
respect to a line paraUe1 to the positive
real axis. The magnitude of each vector
is simply its length on the plot.
In seeking to find the values of which
make the open loop function equal to -1,
the value -1 is considered as a vector
whose angle is 180 degrees ::I:: n 360 degrees, where n is an integer, and whose
magnitude is unity. Then one can consider first the problem of finding the locus
of values for which the angle condition
alone is satisfied. In general, one pictures
the exploratory _ point at various positions on the plane, and imagines the lines
from the poles and zeros to be constructed

so that the angles in turn can be visualized. For any specific problem, however,
many special parts of the locus are established by inspection as illustrated in
examples in later sections. Surprisingly
few trial positions of the , point need be
assumed to permit the complete locus to
be sketched.
After the locus has been determined,
one considers the second condition for a
root, that is. that the magnitude of
KpGp(s)K~G~(s) be unity.
In general,
one selects a particular value of along

the locus, estimates the lengths of the

vectors, and calculates the static gain
KII.K~== l/G,.($)G~(s). After
acquiring
some experience, one usually can select
the desired Position of a dominant root to
detennine the allowable loop gain. The
position of roots along other parts of the
locus usually can be determined with less
than two trials each.
An interesting fact to note from equation 6 is that for very low gain. the roots
are very close to the poles in order that
corresponding vectors be very small. For
very high gain, the roots approach in...
-finity or terminate on a zero.
Single Loop

Example

Consider a single loop system such as

shown in Figure 1 in which the transfer

functions are given in equation 7.

K

a value of just above the real axis. The

decrease in q,o from 180 degrees can be
made equal to the sum of and ~J if the
reciprocal of the length from the trial
point to the origin is equal to the sum of
the reciprocals of lengths from the trial
point to -1/T1 and -lIT,. Ifadamping
ratio of 0.5 for the complex roots is desired, the roots and '. are fixed by the
intersection with the locus of radial lines
at *60 degrees with respect to negative
real axis.
In calculating K for a==rl, it is CODvenient to consider a. term (l+T.a) as a
ratio of lengths from the pole -lIT! to
the point and from _ to the origin respectively. After making gain K-I/[G(s)]...,. a good first trial for finding r.
is to assume that it is near -l/Ta and
solve for (1/T1+.). MtB the roots are
determined to the desired accuracYt the
over-aU transfer functioncan be expressed
as givenin equation 8.

'1

90

~ .. (1-~)( 1-;')(1-;')

(8)

The procedure in handliug a multiple

loop system now can be explained.

Multiple Loop System

Consider a multiple loop system in

which the single loop system just solved

is the forward path of another loop, as
shownin Figure 4.
so/ei is given in factored form by equation 8 so the roots of the inner loop now
serve as base pointsfor the newlocus plot.
For convenience, however, neglect the
eifect of the term (1-./1',,) so that the
locusfor the outer loopis shownin Figure
5.

The locus for the outer loop would be a

circle about the -liT point as a center
if the effect of 8,18, were completely neglected. Actually, the vectors from the
points
and ~t introduce net angles so
that the locusis modifiedas shown. The

'1

Fisul. 2. Root locus plot

110

mum build-up rate, overshoot, natural

'i

frequency of oscillation, and the damping

rate as effective clues in solving this
~------f

FI,ure ~ Caboye). Multiple loop bloclc diI

Fisure 5 (,I,ht).

problem.

--_--..1

1+ T5 .....

Other Applications

9'.'"

Multiple loop root locul

'1

angle at which the locus emerges from

can be found by considering a value of
close to the point n, and solving for
the angle of the vector (.-rt).
Assume that the static loop gain desired is higher than that allowed by the
given system. The first modification
suggested by the plot is to move the
and '1 points farther to the left by obtaining greater damping in the inner loop. If
these points are moved far to the left, the
loci from these points terminate in the
negative real axis and the loci from the
origin curve back and cross the j~ axis.
Moving the -lIT point closer to the
origin would then be effective in permit..
ting still higher loop gain. The next as-

'1

pect of synthesis involves adding corrective networks.

Corrective Networks
Consider a somewhat unusual system.
which arises in instrument servos whose
open loop transfer function is identified
by the poles 1'1 and PI in Figure 6(A). As
loop gain is increased from zero, the roots
which start from PI and PI move directly
toward the unstable half plane. These
roots could be made to move a.way from
the jCIJ axis if 180 degrees phase shift were
added. A simple network to add is three
lag networksin series.each having a time
constant T such that 60 degrees phase
shift is introduced at i'J. The resultant
locus plot is shown in Figure 6(B).
The gain DOW is limited only by the requirement that the new pair of roots do
not cross the jC6J axis. A value of gain is
selected to obtain critical damping of
these roots and the corresponding positions of all the roots are shown in Figures
6(A) and 6(B) as small circles.
Actua.1ly, greater damping could be
achieved for roots which originate at IJs
and PI if a phase shifting bridge were used
rather than the 3-1ag networks. Its
transferfunction is (3- Ts)/(l+Ts) andis
of the "nonminimum phase" type of circuit
Since these types of correction are
somewhat unusual, it is perhaps well to
point out that the analysis has been verified by actual test and application.

These examples serve to indicate the

reasoning process in synthesizing a CODtrol system by root locus method. An
engineer draws upon all of his experience,
however, in seeking to improve a given
system; therefore, it is well to indicate
the correlation between this method and
other methods.

Many systems require a set of simultaneous equations to describe them and

are said to be multicoupled. The corresponding block diagrams have several inputs to each loop so that the root locus
method cannot be applied immediately.
One should first layout the diagram so
that the main line of action of the signals
forms the main loop with incidental
coupling effects appearing as feedbacks
and feed forwards. One then proceeds to
isolate loops by replacing a signal which
comes from within a loop by an equivalent
signal at the output. replacing a signal
entering a loop by an equivalent signal at
the input. One can and should keep the
physical picture of the equivalent system
in mind as these manipulations are carried out.
The techniquesofthe root locusmethod

Correlation with Other Methods

The valuable concepts of frequency response methods- are in a sensemerely extended by the root locus system. Thus
a transfer function with having a complex value rather than just a pure imagi..
nary value corresponds to a damped sinu..
soid being transmitted rather than an un..

can be used effectively in analyzing electric circuits. As a simple example. consider the lead-lag network of Figure
7(A).
It can be shown that the transfer function of this network is as given in equation 9
Vo

(l+R.Ct _) (l+RtCte)R.

~ = (1+R1C1..) (l+RtCt.)Ra+

damped one. The frequency and gain

for which the Nyquist plot passes through
the -1 point are exactly the same values
for which the root locus crosses the j~
uis. Many other correlations appear in

The denominator can be factored algebraically by multiplying out and finding

solving a singleproblemby both methods.

The results ofroot locusanalysiscan be
easily converted to frequency response
data. Thus one merelyassumes valuesof

the zeros of the resulting quadratic. As

an alternative, it win be noted that the
zeros of the denominator must satisfy
equation 10

+ +R.) ct.]

R1 [1 (Rt

(9)

along the jCl) axis, estimates the phase

angles and vector lengths to the zeros and

poles, and calculates the sum of the

angles for total phase shift and the product of lengths for attenuation. The Inverse problem of determining zeros and
poles from experimental data is the more
difficult one. Many techniques are already available, however,such as drawing
asymptotes to the logarithmic attenua..
tion cqrve. For unusual cases, particularly those in which resonant peaks are
involved, the conformal mapping technique originated by Dr. Profos of Swit-

zerlandis recommended. a
The transient response is described by
the poles of the transfer function. The
inverse problemin this caseis to locate the
poles from an experimental transient response. One might use dead time, maxi..
111

Figure 6. (A) Bille .y.tem. (8) Corr.ded

sy.tem

Figur. 7 (I.ft). (A)

Orcult dll,fllll. (B)
RootloclII

FI,ur.

A
I
-I

liA

.:L.
".C,

1-

_0

(1l,+'Y C

I;

(I/RICI+.)(l/R,C,+.)Ra(RICI)(RICs)
(l/(R,+Ra)C,+.}(Rs+Rs)CsRI
--1

(10)

The vectors in this expression are represented according to the root locus
scheme in Figure 7(B). The two roots
are thereby bounded as shown by the two
dots and the cross. Their exact locations
could be estimated or accurately determined by graphical methods.
The locus of roots now is simply in.
tervals along the negative real axis between the open loop zeros and poles as
shownin Figure 7 (B). The exact location
of the roots along these intervals is determined in the usual way. Note that the
constant in equation 10 is of the form
R'Ch in which R' is the effective value of
R1 and R. in parallel.
In more complicated networks, the advantages of the root locus concept over
algebraic methods becomes greater; its
particular advantage is in retaining at all
times II clear picture of the relationships
between the over-all network parameters
and the parameters of individual circuit
elements.
In the classical problem of finding
roots, the differential equation is given in
the form of a sum of terms of successively
higher order. This can be converted to
the form shown in equation 10

'''+0'''-'+'''''-'+ ... +m

- ('+0).+11]'+ . +m

(11)

This corresponds to a block diagram

with another loop closed for each higher

8 (rlg"t).
Spirul.

order term. Solve for the roots of the

first loop which corresponds to the quantities in brackets above and proceed as
before for the multiple loop system. If
the roots close to the origin are of most
interest, substitute .-1/" first and solve
for root values of x, Other combinations
are, of course, possible because a single
root locusbasically determines the factors
of the sum of two terms.
The root locus method is thus an analytical tool which can be applied to
other problems than control sy.stl:m synthesis for. which it was developed. But in
attacking a new problem one would probably do' well to try first to develop a
method of analysis which is natural for
that problem rather than seek to apply
any existing methods.

Graphical Calculations
The root locus plot is first established
in approximate form by inspection. Any
significant point on the locus then can be
checked by using the techniques indicated
in this section. Note that only two calculations llR involved, adding angles and
multiplying lengths. Fortunately, all of
these angles and lengths can be measured
at the point. Thus angles previously
pictured at the zeros and poles also appear at the point but between a borizontal line to the left and lines to the
zeros and poles. A piece of transparent
paper or plastic pivoted at the. point can
be rotated successively through each of
these angles to obtain their sum.
The reader can duplicate the "spirulc"
with two pieces of transparent paper, one
for the disk and the other for the arm.

No Discussion

112

Several procedures are possible, but the

over-all. purpose is to successively rotate
the arm with respect to the disk through
each of the angles of interest. Thus for
adding phase angles, the disk is held
fixed while the arm is rotated from a pole
to the horizontal, whereas the two move
together in getting aligned on the next
pole. For multiplying lengths, the disk is
held fixed while the arm is rotated from
the position where II. poleis on the straight
line to the position where the pole is on
the lo~arithmic curve. Rotations are
made in the opposite directions for zeros
than they are for poles.

Conclusions
The definite opinion of engineers using
this method is that its prime advantage
is the complete picture of a system which
the root locus plot presents . Changing
an open loop parameter merely shifts a
point and modifiesthe locus. By means of
the root locus method, all of the zeros and
poles of the over-all function can be determined.
Any linear system is completely defined by this determination, and its response to any particular input function
can be determined readily by standard
mathematical or graphical methods.

References
1

PRlNc:Jft,SI

2.

TIlANllaHU

o.

SSlwoNaC:RANlSMI

(book).

SYS~.NI

(hoole),

G. S. Browll. D. P. ClmpbeU. Joh .. Wiley IDd

Sou. Ne" York. N. Y 19.5.
lH

LINSAI

14.1'. GUolIIlf,J. L.BI...e Johll WUey lad So....

N." York, N. Y., 11142.
3.

GllUmCAL ANAt.YIII O. Comao&. SYITSMI.

W. Jl. IYIII..

AlEE TrollJlUliou. volume 117.

lllU. P"I" 647-51.